Abstract
We introduce a class of FBI transforms using weight functions (which includes the subclass of Sjöstrand’s FBI transforms used by Christ in (Commun Partial Differ Equ 22(3–4):359–379, 1997)) that is well suited when dealing with ultradifferentiable functions (see Definition 2.3) and ultradistributions (see Definition 2.15) defined by weight functions in the sense of Braun, Meise and Taylor (BMT). We show how to characterize local regularity of BMT ultradistributions using this wider class of FBI transform and, as an application, we characterize the BMT vectors (see Definition 1.2) and prove a relation between BMT local regularity and BMT vectors.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The purpose of this paper is twofold: (i) to explore a new class of FBI transform and show that it can be used to characterize regularity in the classes of ultradifferentiable functions in the sense of Braun et al. [12]; and (ii) use these techniques to study regularity of iterated hypoelliptic constant coefficient partial differential operators. This new systematic approach has led us to a plethora of unanticipated results. We show that the FBI transforms introduced in (i) are not only fundamental to obtain the results in (ii) but also allows us to extend similar results that appeared recently where a similar program was developed with the use of the Fourier transform.
1.1 The Augmented Class of FBI Transforms
The Fourier transform can be used to characterize smoothness of distributions (Paley–Wiener Theorem) and also can be used to characterize analyticity of distributions. However, the analyticity characterization is significantly more difficult than the smoothness characterization (see [22]). An alternative tool to characterize regularity (smooth and analytic) is the FBI transform (see [4]).
In the next paragraphs we will recall several variations and generalizations of the classical FBI transform. Given \(n \in \mathbb {N}\) and \(0<\tau \le 1\) consider the following form in \(\mathbb {R}^n \times \mathbb {R}^n\)
where \(\xi =(\xi _1, \dots , \xi _n)\), \(x=(x_1,\dots , x_n)\) and \(\langle \xi \rangle =\sqrt{1+\sum _{j=1}^n \xi _j^2}.\) Define the function \(a_{\tau }: \mathbb {R}^n \times \mathbb {R}^n \rightarrow {{\mathbb {C}}}\) with the property that
Using the above form in [13] M. Christ defined the following variation of the FBI transform (in the original article \(\tau \) is actually \(\lambda \))
where \((x,\xi )\in \mathbb {R}^n \times \mathbb {R}^n\), \(u\in \mathcal {E}'(\mathbb {R}^n)\) (i.e., u is a compactly supported distribution) and the pairing refers to the interaction between distributions and test functions with respect to the variable \(x'\in \mathbb {R}^n\). Moreover, among other interesting results he proved that for a given \(s>1\) a distribution \(u\in \mathcal {D}'(\mathbb {R}^n)\) is \(G^s\) (Gevrey of order s, see next paragraph) in a neighborhood of \(x_0\) if and only if there exist \(\tau \ge \frac{1}{s}\), \(v\in \mathcal {E}'(\mathbb {R}^n)\) with \(v\equiv u\) in a neighborhood of \(x_0\), positive constants a, C and an open neighborhood V of \(x_0\) such that
Observe that the “limit" choice of \(\tau \) to study \(G^s\) regularity is \(\tau = 1/s\).
In [18] we considered a more general class of FBI transform first introduced in [3] (see also [2] where this FBI transform plays a fundamental role) and we showed that it can be used to characterize Denjoy–Carleman regularity as we now explain. Consider a positive sequence \(M=(M_j)\) satisfying some special properties. If \(U \subset \mathbb {R}^n\) is an open set and \(f\in C^{\infty }(U)\) we say that f is in \({\mathcal {E}}^M(U)\) (f is M-Denjoy–Carleman in U) if for each compact set \(K \subset U\) there exist positive constants C, h such that \(|\partial ^{\alpha } f(x)| \le Ch^{|\alpha |}M_{|\alpha |},\) for each \(x \in K\) and \(\alpha \in \mathbb {N}_0^n.\) The space of compactly supported functions in \(\mathcal {E}^M(U)\) is denoted by \(\mathcal {D}^M(U).\) We equip \(\mathcal {E}^M(U)\) and \(\mathcal {D}^M(U)\) with their usual topologies, the topological duals of these spaces are denoted by \({\mathcal {E}^M}'(U)\) and \({\mathcal {D}^M}'(U)\) respectively (see [26] for more information). The Gevrey spaces \(G^s\) of order s are given by choosing \(M=(j!^s)\). For a fixed sequence \(M=(M_j)\) its associated function is defined by \(M(t)\doteq \sup _j \log (t^j/M_j)\).
The main result in [18] can now be stated as follows: for u in \({\mathcal {E}^M}'(U)\), \(0<\tau \le 1\) and \(k\in \mathbb {N}\) denote
Then u is \( \mathcal {E}^M\) in a neighborhood of \(x_0\) if and only if there exist \(0<\tau \le 1\) so that \(M(t)=O(t^{\tau })\) as \(t\rightarrow \infty \), \(v\in {\mathcal {E}^M}'(\mathbb {R}^n)\) with \(v\equiv u\) in a neighborhood of \(x_0\), positive constants \(a,C>0\) and an open neighborhood V of \(x_0\) such that
A natural question that arises is the following: is there a “limit" choice of \(\tau \)? When dealing with the Gevrey class of order s it follows that the choice is \(\tau =1/s\). However this is far from being trivial when M is not a Gevrey sequence. Note that the choice \({\tau } \ge \frac{1}{s}\) is equivalent to the inclusion \(G^{1/\tau }\subset G^{s}\).
Since \(m(t)=t^{\tau }\) is equivalent to the associated function M of the sequence \((M_j)=(j!^{1/\tau })\), meaning there exist constants \(C_1,C_2\) and \(a>0\) such that \(C_1M(t)\le m(t) \le C_2M(t),\) for all \(t>a\) (consequently, \(\mathcal {E}^M=\mathcal {E}^m\) as per Definition 2.3), a naive approach to try to answer this question is to enlarge the class of the FBI transform considered by allowing the term \(\Vert \xi \Vert ^\tau \) in (1.1) to be any possible weight function as given in Definition 2.1.
In order to justify the previous sentence we need to recall the definition of ultradifferentiable functions by means of weight functions as introduced in [12] (see Definitions 2.1 and 2.3).
Recall that, given two weight functions \(\omega \) and \(\sigma \) the condition \(\omega (t)=O(\sigma (t))\) as \(t\rightarrow \infty \) implies the inclusion \(\mathcal {E}^{\sigma }(\Omega )\subset \mathcal {E}^{\omega }(\Omega )\), for any open set \(\Omega \subset {\mathbb {R}}^n\).
Moving on, given any ultradifferentiable class \({\mathcal {E}}^\omega \) as in Definition 2.3 the augmented class of FBI transforms announced earlier which will be suitable to study regularity problems in \({\mathcal {E}}^\omega \) would be those allowing weight functions \(\sigma (\Vert \xi \Vert )\) instead of \(\Vert \xi \Vert ^\tau \) in (1.2) as long as \(\omega (t)=O(\sigma (t))\) as \(t\rightarrow \infty \). However, the lack of regularity on general weight functions \(\sigma \) make it very difficult to deal with these FBI transforms. In order to avoid this problem we will prove that for each weight function there exists an equivalent one possessing the desired regularity (see Definition 2.18 for the precise meaning of equivalence used here) and therefore define the same class of ultradifferentiable functions, see Remark 2.19.
To be more precise given a weight function \(\sigma \) there exists \(\mu _{\sigma } \in C^1((1,\infty ))\) such that \(\sigma \) and \(\mu _{\sigma }\) are equivalent (see Proposition 2.20). Let \(\kappa >0\), following [13] we consider the differential form in \(\mathbb {R}^n \times \mathbb {R}^n\)
and the function \(a^{\kappa }_{\mu _{\sigma }}:\mathbb {R}^n \times \mathbb {R}^n\rightarrow {{\mathbb {C}}}\) defined by the property
Note that, fixing \(\xi \in \mathbb {R}^n\) it follows that \(x' \mapsto a^{\kappa }_{\mu _{\sigma }}(x',\xi )\) is a polynomial function (consequently it is a \(\mathcal {E}^{\omega }\) function). Hence, for each \(u \in {\mathcal {D}^{{\sigma }}}'({\mathbb {R}}^n)\) and \(\phi \in {\mathcal {D}^{{\sigma }}}({\mathbb {R}}^n)\) we can define
Here the notation \(u_{x'}\) is to emphasize that the ultradistribution u is acting on the function in \(x'\) and the other variables are thought as parameters. When \(\kappa =1\) we simply denote \({{\mathcal {F}}}_{\mu _{\sigma }}={{\mathcal {F}}}^{\kappa }_{\mu _{\sigma }}.\)
Our main result is the following FBI characterization of ultradifferentiable functions that can be viewed as a Paley-Wiener type theorem.
Theorem 1.1
Fix a weight function \(\omega \) (see Definition 2.1), \(\Omega \subset {\mathbb {R}}^n\) an open set and \(u\in {\mathcal {D}^{\omega }}' (\mathbb {R}^n)\). In order that \(u \in \mathcal {E}^{\omega }\) in a neighborhood of \(x_0\in \Omega \) it is necessary and sufficient that there exist a weight function \(\sigma \) with \(\omega (t)=O(\sigma (t))\) as \(t \rightarrow + \infty \) so that for each \(\phi \in \mathcal {D}^{\sigma }(\Omega )\) there exist \(C,c>0\) and a neighborhood \(V\subset \Omega \) of \(x_0\) such that
To prove Theorem 1.1 we actually show a slightly stronger result, see Theorem 4.2.
As it is customary in these Paley-Wiener type results one of the main ingredients to prove the sufficient part of Theorem 1.1 is the inversion formula of the FBI transform. We provide two different versions for the FBI inversion formula, see Lemmas 5.3 and 5.5. It turns out that we use the first inversion formula, Lemma 5.3, to prove the second Lemma 5.5 and this was inspired by [32, Lemma IX.4.1].
Observe that using [12, 8.9 Remark] it follows that for each sequence \((M_j)\) considered in [18] there exists a function \(\omega _M\) satisfying Definition 2.1 such that the space of ultradifferentiable functions defined by \((M_j)\) coincides with the space defined by \(\omega _M.\) However, given a function \(\omega \) satisfying Definition 2.1 it is necessary to impose on \(\omega \) an additional stronger condition (the existence of a constant \(H>0\) such that \(2\omega (t) \le \omega (Ht) + H,\) for each \(t>0\)) to obtain a sequence \((M_j)\) such that \(\mathcal {E}^{M}= \mathcal {E}^{\omega }\) (see [11]). Hence, for general weight functions one can think that the local regularity characterization results presented here as natural extensions from the ones given in [18].
Previous Paley–Wiener type results for ultradifferentiable functions appeared in the literature but with the use of the Fourier transform, [12], and this is an obstacle to work with elements in their dual. The use of FBI transforms allow us to work also with ultradistributions, the natural ambient.
1.2 Application to Ultradifferentiable Vectors
As an application of Theorem 1.1 we study BMT vectors (or iterates) of constant coefficients partial differential operators \(P(D)=\sum _{|\alpha |\le m}a_\alpha D^\alpha \). To be more precise recall that in [29] Nelson introduced the set of analytic vectors of a partial differential operator with analytic coefficients P(x, D) in an open set \(U\subset \mathbb {R}^n\) and proved that analytic vectors are real analytic functions exactly when P(x, D) is elliptic.
Later Komatsu, in [25], as well as Kotake and Narasimhan, in [27], obtained a slightly improvement of Nelson’s result as follows: let \(U \subset \mathbb {R}^n\) and P(x, D) as before then a function \(f\in L^2_{loc}(U)\) is real analytic in U if and only if
-
1.
\(P^jf \in L^2_{loc}(U)\) (in the sense of the distributions), for each j, and
-
2.
\(\forall K \subset U\) compact there exist \(C,h>0\) such that
$$\begin{aligned} \Vert P^j f \Vert _{L^2(K)} \le C h^j (h j)^{jm}, \quad j \in \{0,1, \dots \}. \end{aligned}$$(1.6)
Here and throughout these notes \(P^j f \doteq (P\circ \cdots \circ P) f\), for \(j\in {\mathbb {N}},\) and \(P^0 f \doteq f\).
Newberger and Zielezny initiated an investigation in the Gevrey category \(G^s\) (replacing \(j!^m\) by \(j!^{sm}\) in (1.6)), see [30]. It is worth mentioning the work of Baouendi and Métivier [1] that deals with the case when P(D) is of principal type and hypoelliptic with analytic coefficients. They observed that if u is an s-Gevrey vector of P (in a smaller neighborhood) then there exist \(s'>s\) such that u is \(s'\)-Gevrey. There is a vast literature concerning optimal regularity for such Gevrey vectors, \(G^s\), \(s\ge 1\), for instance [15, 16, 31] to cite just a few. Also, for elliptic operators a proof of the Kotake–Narasimhan theorem in some classes of ultradifferentiable functions is given in [7].
Recently, regularity problems for ultradifferentiable vectors defined by weight functions in the sense of Braun et al. [12] for constant coefficients operators (see Definition 1.2) appeared in the literature. It turns out that completeness of these spaces is equivalent to the hypoellipticity of P [23]. Additionally, there are others characterizations in terms of the decay of the Fourier transform, [5, 8, 24]. It is worth mentioning that in all of these results, the authors prove Paley–Wiener theorems only for distributions when the natural object to deal with are ultradistributions (see implication \((2) \Rightarrow (1)\) at Theorem 1.3). We define BMT-vectors as follows:
Definition 1.2
For each weight function \(\omega \), open set \(\Omega \subset \mathbb {R}^n\) and polynomial function \(P(\xi )\) of degree m we define \(\mathcal {E}^{\omega } (\Omega ;P)\) as the set of \(u\in C^{\infty } (\Omega )\) such that for each compact set \(K \subset \Omega \) there exists \(\lambda >0\) satisfying
where we are denoting \( \varphi ^*(x)=\varphi _{\omega }^*(x) \doteq \sup \{xy-\varphi (y): \, y \ge 0\},\) for each \(x>0\) (see (2.1)).
We will call the elements of \(\mathcal {E}^{\omega } (\Omega ;P)\) BMT vectors.
To state our main result for ultradifferentiable vectors we need to introduce some ingredients. Let \(P(D)=\sum _{|\alpha |\le m}a_\alpha D^\alpha \) be a hypoelliptic constant coefficients partial differential operator. Then, it is well known (see [21]) that there exist constants \(C,K>0\) and a positive number \({\tilde{\rho }}\le 1\) such that
Define the hypoelliptic index by
It is known that this supremum is attained and it is a rational number. In other words there exist constants \(C,K>0\) such that
We will refer to the number \(\rho \) as the Hörmander hypoellipticity index of P(D). Is is important to mention that this condition also plays an important role in [10]. For a fixed weight function \(\omega \) we will define \(\sigma (t)=\sigma _{\omega ,\rho }(t)=\omega (t^{\rho }).\) Note that, if \(\omega \) is a weight function and \(0<\rho \le 1\) then \(\sigma \) is a weight function.
The next result gives a characterization of the spaces \(\mathcal {E}^{\omega }(\Omega ;P)\) (see Definition 1.2) in terms of the FBI transforms introduced here. The characterization of vectors of partial differential operators using Fourier transform was first given in [9].
Theorem 1.3
Fix a weight function \(\omega \), a hypoelliptic constant coefficient partial differential operator P(D) of order m on an open set \(\Omega \subset {\mathbb {R}}^n\) along with his hypoellipticity index \(\rho \) and define the weight function \(\sigma \) as before. Let \(\varphi (t)=\omega (e^t)\) and \(\varphi ^*\) is Young conjugate, see (2.1). Given \(u \in {\mathcal {D}^{\sigma }}' (\Omega )\) and \(x_0\in \Omega \) the following conditions are equivalent:
-
1.
There exists a neighborhood U of \(x_0\) such that \(u \in \mathcal {E}^{\omega }(U;P)\).
-
2.
There exist \(\phi \in \mathcal {D}^{\omega }(\mathbb {R}^n)\) (such that \(\phi \equiv 1\) in a neighborhood of \(x_ 0\)), \(C,\lambda , c>0\) and a neighborhood V of \(x_0\) such that
$$\begin{aligned}{} & {} |\mathcal {F}_{\mu _{\sigma }} \left( \phi \, [P(D)]^{N} ( u)\right) (x,\xi ) | \le C e^{\frac{1}{\lambda } \varphi ^*(Nm \lambda )} e^{-c \sigma (\Vert \xi \Vert )},\nonumber \\ {}{} & {} \quad (x,\xi ,N) \in V \times \mathbb {R}^n \times \mathbb {N}_0. \end{aligned}$$(1.9)
One of the main tools to prove Theorem 1.3 is a version of [20, Theorem 4.1] specialized to balls on which we obtain a better estimate (see Lemma A.1). This allows us to get rid of a strong restriction on the weight functions treated in [6, Theorem 3.3]. Specifically, in their theorem they consider weight functions \(\omega \) so that \(\omega \left( t^{\gamma }\right) =o(\sigma (t))\), as \(t \rightarrow +\infty \), where \(\gamma \) is a constant that arises from the hypoellipticity of P and \(\sigma (t)=t^{1 / s}\) for some \(s>1\). This is equivalent to \(\omega \left( t\right) =o(t^{1/\gamma s})\), as \(t \rightarrow +\infty \) which in turn, implies that \(G^{s\gamma }\subset {\mathcal {E}}^\omega \).
Another important improvement is that we allow \(u \in {\mathcal {D}^{\sigma }}' (\Omega )\) while in [6, Theorem 3.3] the authors considered only \(u \in \mathcal {D}' (\Omega )\). Note that both, inequality (1.5) in Theorem 1.1 and condition (2) in Theorem 1.3 can be checked if u is only an ultradistribution and this will be used in a forthcoming paper to define wave front-sets for ultradistributions.
Another application is given in Corollary 6.4 where we recover the Kotake-Narasimhan theorem in this context. Also, it is important to recall that similar results for global Gevrey classes were studied in [19].
In the following we will use the notation B(x, r) to denote the ball of radius \(r>0\) and centered at \(x\in {\mathbb {R}}^n\). We will also use the notation \({\mathbb {N}}_0^n\) to denote the set of all multiindices \(\alpha =(\alpha _1, \dots , \alpha _n)\) with \(\alpha _j\in \{0,1,2,\dots \}\) for each \(j\in \{1,\dots ,n\}\) and denote \(|\alpha |=\alpha _1+ \cdots + \alpha _n.\)
The organization of this paper is as follows: in the second chapter we recall the definitions of weight functions, ultradifferentiable functions and ultradistributions as well as we introduce the function \(\mu _{\omega }\). We prove the necessity of Theorem 1.1 in chapter 3. In chapter 4 we present inversion formulas of our class of FBI transforms and we use them to prove the sufficiency of Theorem 1.1. A characterization of iterates of constant coefficients operators and a relation between ultradifferentiable functions and iterates of constant coefficients operators using our class of FBI transforms is given in chapter 5.
2 Ultradifferentiable Functions Defined by Weight Functions
This section is devoted to recalling basic definitions concerning ultradifferentiable functions and ultradistributions. First we establish the concept of weight functions used throughout this work.
Definition 2.1
A continuous function \(\omega : [0, +\infty [ \rightarrow [0,+ \infty [ \) with \(\omega \equiv 0\) in [0, 1] and increasing in \([0,+\infty )\) is called a weight function when the following conditions are satisfied:
Moreover, given \(x=(x_1, \dots , x_n) \in \mathbb {R}^n\) we will write \(\omega (x)=\omega \left( \left( \sum _{j=1}^n |x_j|\right) ^{1/2} \right) .\)
Example 2.2
For each \(s>1\), let \(\omega _s \equiv 0\) in [0, 1] and \(\omega _s(t)=t^{\frac{1}{s}}-1 \), for \(t>1\). Note that, \(\omega _s\) is a weight function.
2.1 Ultradifferentiable Functions
Before we introduce the notion of ultradifferentiable functions we need to recall the notion of Young conjugate that will be used throughout this paper. Let \(\omega \) be a weight function and let \(\varphi \) be defined by (\(\delta \)). The Young conjugate of \(\varphi \), \(\varphi ^*\), is defined by
It is important to recall that, by \(\gamma \)) and \(\delta \)), the Young conjugate of \(\varphi \) is well defined. Moreover, \(\varphi ^*\) is an increasing convex function, \(\varphi ^*(0)=0,\) \(\lim _{t \rightarrow + \infty } \frac{\varphi ^*(t)}{t} = + \infty \) and \((\varphi ^*)^*=\varphi \) (see [12, 1.3 Remark]).
Definition 2.3
Let \(K \subset \mathbb {R}^n\) be a regular compactFootnote 1 set and \(C^{\infty } (K)\) as in [26]. For each weight function \(\omega \) and \(\lambda >0\) we define
and
Let \(\Omega \) be an open subset of \({\mathbb {R}}^n\), we also define
Remark 2.4
Let \(\Omega \) be an open subset of \(\mathbb {R}^n\). It follows that \(\mathcal {E}^{\omega }(\Omega )\) is the set of smooth functions \(f:\Omega \rightarrow \mathbb {C}\) such that for each compact set \(K\subset \Omega \) there exist \(\lambda , C>0\) satisfying
And \({{\mathcal {D}}}^{\omega } (\Omega )\) is the subset of \(\mathcal {E}^{\omega }(\Omega )\) consisting of all compactly supported functions.
Remark 2.5
For each open set \(\Omega \subset {\mathbb {R}}^n\) condition (\(\beta \)) guarantees that \({{\mathcal {D}}}^{\omega } (\Omega ) \ne \{0\}\) (see [12, 2.5 Corollary]).
Example 2.6
If \(s>1\) and \(\omega _s\) is given as in Example 2.2, then the space \(\mathcal {E}^{\omega _s}(\Omega )\) coincides with the Gevrey class of order s, \(G^s(\Omega )\), for each open set \(\Omega \subset {\mathbb {R}}^n\).
2.2 Technical Results on Weight Functions
This subsection is dedicated to the presentation of some technical results on weight functions and its Young conjugate. In the end of this subsection we will prove an important characterization of ultradistributions.
Throughout this section we will consider a fixed weight function \(\omega \) (see Definition 2.1).
Remark 2.7
Using (\(\gamma \)) we see that there exists \(A>1\) such that
Thus, applying (\(\alpha \)) it follows that
Next we will present some properties of \(\varphi ^*\) (see (2.1)).
Remark 2.8
Note that, \(\varphi ^*\) is an increasing function, is a convex function, \(\varphi ^*(0)=0,\) \(\lim _{t \rightarrow + \infty } \frac{\varphi ^*(t)}{t} = + \infty ,\) \(\frac{\varphi ^*(t)}{t}\) is increasing and \((\varphi ^*)^*=\varphi \) (see [12, 1.3 Remark]).
Thus,
Remark 2.9
Using condition (\(\alpha \)) and [12, 1.4 Lemma] it follows that,
Thus, for any \(\lambda ,\alpha >0\) choosing \(t= \alpha \lambda L^k\) we obtain
Note that for each \(h>0\) there exists \(k \in \mathbb {N}\) such that \(\log h<k\). Thus, considering \(\alpha =j\) (for any \(j \in \mathbb {N}\)), it follows that
Thus, denoting \(C_{h}=\exp \{\frac{1}{ L^k} \sum _{q=1}^kL^q\}\) and \(\lambda _{h}=\lambda L^k\), we conclude that
for each \(h,\lambda >0\) and \(j\in \mathbb {N}.\)
Remark 2.10
If \(\omega \) and \(\sigma \) are weight functions such that there exists \(A,C>0\) satisfying
then there exists constants \(\lambda _{\sigma }, {\tilde{L}}>0\) such that \(\varphi _{\sigma }^*(t)\le \frac{1}{\lambda _{\sigma }}\varphi _{\omega }^*(\lambda _{\sigma }t)+{\tilde{L}}\), \(\forall t\ge 0\).
In order to prove the above claim we will consider \(L>1\) such that \(\sigma (ex) \le L(\sigma (x)+1),\) for each \(x>0.\) In particular,
for each \(m \in \mathbb {N}\) and \(x>0.\) Thus, considering m such that \(e^m>A\)
for \(\lambda _{\sigma }=L^mC\) and for each \(t>0.\)
Remark 2.11
Since \(\omega \) is increasing it follows that,
hence, using (\(\beta \)), it follows that \(\omega (t)=o(t)\) (\(t \rightarrow + \infty \)).
Remark 2.12
Using the definition of \(\varphi \) and \(\varphi ^*\), for each \(b \ge 1\) and \(\lambda >0\) we have,
Thus, considering \(x\ge 1\) and \(y=\log x \ge 0,\)
Moreover, since \(\omega (x)=0\) (when \(x \in [0,1]\)), (2.7) also follows for \(x \in [0,1].\) Furthermore, choosing \(b=x\), it follows that, \( b^b \le e^{\frac{1}{\lambda } \varphi ^*(\lambda b) +\frac{1}{\lambda } \omega (b)}. \) Thus, using Remark 2.11, there exists \(b^*>0\) (independent of \(\lambda \)) such that
Therefore, for each \(\lambda >0\), there exists \(C_{\lambda }\) such that
Remark 2.13
Using (2.8) and (2.5) it follows that the space of all real analytic functions is contained in \(\mathcal {E}^{\omega }.\)
Remark 2.14
Observe that, using (2.7), for \(b=n+1\), it follows that
for each \(a,c>0\) and \(r\ge 1.\)
2.3 Ultradistributions
Definition 2.15
The continuous dual of \( \mathcal {D}^{\omega }(\Omega )\) is denoted by \({\mathcal {D}^{\omega }}^{\prime }(\Omega )\) and its elements are called ultradistributions. Also, the continuous dual of \({\mathcal {E}^{\omega }}(\Omega )\) is denoted by \({\mathcal {E}^{\omega }}^{\prime }(\Omega )\).
Remark 2.16
It follows from the definition that a linear functional \(u:\mathcal {D}^{\omega }(\Omega ) \rightarrow \mathbb {C}\) is continuous if and only if u is sequentially continuous.
Lemma 2.17
Let \(u: \mathcal {D}^{\omega }(\Omega ) \rightarrow \mathbb {C} \) be a linear functional. The following statements are equivalent:
-
1.
u is (sequentially) continuous;
-
2.
For all compact \(K \subset \Omega \) and for each \(\delta >0\) there exists a real positive constant \(C_{\delta ,K}\) such that
$$\begin{aligned} \left| \langle u, \phi \rangle \right| \le C_{\delta ,K} \sum _{\alpha \in \mathbb {N}_0^{n}} e^{-\delta \varphi ^*(|\alpha |/\delta )} \Vert \partial ^{\alpha } \phi \Vert _{L^{\infty }(K)}, \quad \quad \forall \phi \in \mathcal {D}^{\omega }(K) \doteq \mathcal {E}^{\omega }(\Omega ) \cap C_0^{\infty }(K).\nonumber \\ \end{aligned}$$(2.10)
Proof
Consider u such that (2.10) is satisfied. If \(\phi _j \rightarrow 0\) in \(\mathcal {D}^{\omega }(\Omega )\), then there exists a compact set \(K\subset \Omega \) such that \(\text{ supp } \, \phi _j \subset K\) (for all \(j \in \mathbb {N}\)) and there exists \(\lambda >0\) such that
Hence, for each \(\epsilon >0\) there exists \(j_\epsilon \in \mathbb {N}\) such that
when \(j>j_{\epsilon }.\) Thus, for each \(\delta >0\) there exists \(C_{\delta ,K}>0\) such that,
where \(j>j_{\epsilon }\). Considering \(\delta =1/(\lambda L) \) (where L is the constant considered in (\(\alpha \)) and using (2.4) it follows that
where \(A=A( \lambda )=\frac{1}{\lambda } \). Hence,
Since \(\sum _{\alpha \in \mathbb {N}_0^n} e^{-|\alpha |}<+\infty \), we can conclude that \(\langle u, \phi _j \rangle \rightarrow 0.\) Therefore, u is continuous.
Conversely, suppose by contradiction that there exist a compact set \(K \subset \Omega \) and \(\delta >0\) such that for each \(C>0\) there exists \(\phi =\phi _C \in \mathcal {D}^{\omega }(K)\) where
Thus, for each \(j \in \mathbb {N}\) there exists \(\phi _j \in \mathcal {D}^{\omega }(K)\) such that
Set \(\psi _j= \phi _j/r_j.\) Note that \(\text{ supp } \, \psi _j \subset K\) and \(e^{-\delta \varphi ^*(|\alpha |/ \delta )} \Vert \partial ^{\alpha } \psi _j \Vert _{L^{\infty }(K)} \le \frac{1}{j} \) (\(\forall \alpha \in \mathbb {N}_0^n\)). Hence, \( {\psi }_j \rightarrow 0\) in \(\mathcal {D}^{\omega }(\Omega )\) and \(|\langle u, \psi _j \rangle |=1.\) This fact means that a contradiction was obtained. Therefore, u is not continuous. \(\square \)
2.4 A Regular Equivalent Weight Function
Definition 2.18
Let \(\omega :[0,+\infty [ \rightarrow [0,+\infty [\) and \(\rho :[0,+\infty [ \rightarrow [0,+\infty [\) be functions. We say that \(\omega \) and \(\rho \) are equivalent (\(\omega \sim \rho \)) when there exist \(\delta ,A,C>0\) such that \(A\omega (t) \le \rho (t) \le C \omega (t),\) for \(t>\delta .\)
Remark 2.19
Let \(\omega :[0,+\infty [ \rightarrow [0,+\infty [\) and \(\rho :[0,+\infty [ \rightarrow [0,+\infty [\) be weight functions. If \(\omega \sim \rho \) then \({{\mathcal {E}}}^{\omega }={{\mathcal {E}}}^{\rho }\) and \({{\mathcal {D}}}^{\omega }={{\mathcal {D}}}^{\rho }\) (see [12, 3.4 Proposition]).
The next proposition guarantees that for each weight function \(\omega \) there exists a \(C^1\) function \(\mu \) such that \(\omega \sim \mu \). Therefore, we recall that by (2.2) there exist \(A,H>1\) such that \(\omega (et)\le H\,\omega (t),\) when \(t\ge A.\) Fix \(h> \max \{ 2;\log (He)\}.\)
Proposition 2.20
There exists a function \(Q \in C^{1}((1, +\infty ))\) satisfying:
-
1.
\(\omega \sim Q\) and \(\displaystyle \lim _{y \rightarrow + \infty } Q'(y)=0.\)
-
2.
There exist \(\delta _{\omega } >0\) and \(D>0\) such that, if \(\mu _{\omega }(y) \doteq (h-1)Q \left( \left( \delta _{\omega } + \Vert y\Vert ^2\right) ^{1/2}\right) , \) for \(y \in \mathbb {R}^n,\) then,
$$\begin{aligned} \mu _{\omega } (\xi )\ge & {} \omega \left( \left( \delta _{\omega } + \Vert \xi \Vert ^2\right) ^{1/2}\right) >0, \quad \forall \xi \in \mathbb {R}^n; \end{aligned}$$(2.11)$$\begin{aligned} \mu _{\omega } (\xi )\le & {} D \,\omega ((\delta _{\omega }+\Vert \xi \Vert ^2)^{1/2}), \quad \forall \xi \in \mathbb {R}^n; \end{aligned}$$(2.12)$$\begin{aligned} |\partial _{\xi _j} \mu _{\omega }(\xi )|< & {} 1, \quad \quad \quad \quad \quad \quad \quad \quad \quad \forall \xi \in \mathbb {R}^n \,\, \text{ and } \,\, j \in \{1,\dots ,n\}. \end{aligned}$$(2.13) -
3.
Denoting \(a^{\kappa }_{\mu _\omega }\) as in (1.3) it follows that for each compact set \(K\subset \mathbb {R}^n\) there exists \(C_K>0\) (independent of \(\xi \)) such that \(\sup _{x\in K}|\partial _x^\alpha \{ a^{\kappa }_{\mu _\omega }(x,\xi )\}| \le C_{K},\) for each \(\xi \in \mathbb {R}^n\) and \(\alpha \in {\mathbb {N}}^n_0\).
Proof
1. Define
Since \(h>2\) it follows that \(0\le \frac{\omega (s)}{s^h}\le \frac{\omega (s)}{s^2},\) when \(s>1\). Recalling that \(\omega \equiv 0\) in [0, 1] and using (\(\beta \)) we can see that the function Q is well defined. And
Therefore, \(Q'\) is a continuous function. Moreover, since \(\omega \) is increasing and \(h>2\) we have
And, since \(h> \log (He)\) we obtain \(e^h/(He)>1.\) Consequently, \(C=\sum _{j=0}^{+ \infty } \left( \frac{He}{e^{h}}\right) ^j<+\infty .\) Thus,
when \(y\ge A.\) Moreover, recalling that \(h > 2\)
Thus using (\(\beta \)) and (2.6) it follows that \(\lim _{y \rightarrow + \infty } |Q'(y)|=0.\)
2. Since \(\lim _{y \rightarrow + \infty } |Q'(y)|=0\) there is \(\delta _{\omega }>A\) such that \(|Q'(y)|<\frac{1}{h-1}\) when \(y>\delta _{\omega }^{1/2}.\) Hence, if \(\mu _{\omega }\) is as being defined above then, \(|\partial _{\xi _j} \mu _{\omega }(\xi )|<1,\) for each \(\xi \in \mathbb {R}^n\) and \(j \in \{1, \dots , n\}.\) Moreover, using (2.14) it follows that \( \mu _{\omega } (\xi )=(h-1)Q \left( \left( \delta _{\omega } + \Vert \xi \Vert ^2\right) ^{1/2}\right) \ge \omega \left( \left( \delta _{\omega } + \Vert \xi \Vert ^2\right) ^{1/2}\right) , \) \(\forall \xi \in \mathbb {R}^n. \) And, using (2.15) we obtain \( \mu _{\omega } (\xi )\le D \,\omega ((\delta _{\omega }+\Vert \xi \Vert ^2)^{1/2}) \) where \(D=(h-1)CH(e-1). \)
3. It follows from (2.13) and the definition of \(a^{\kappa }_{\mu _\omega }.\) \(\square \)
Remark 2.21
Since \(\omega \) satisfies \(\delta \)), it follows that the Q function defined in the proof of Proposition 2.20 also satisfies \(\delta \)).
Remark 2.22
The Q function defined in the proof of Proposition 2.20 was inspired by the proof of [28, 1.3. Proposition]. The function presented there is \(\chi (y) = \int _1^{+\infty } \frac{\omega (yt)}{t^2} \, dt\). In their work, the authors establish a similar inequality to (2.15) by assuming a stronger hypothesis than ours: the existence of \( K > 1 \), \( T > 0 \), and \( 0< \epsilon < 1 \) such that \(\omega (Kt) \le (K - \epsilon ) \omega (t)\) for each \( t \ge T \). However, since we do not have this hypothesis, we considered the power h instead of 2.
3 Technical Results on FBI Transforms
This chapter is dedicated to the presentation of some technical results. Using Faà di Bruno’s formula we will obtain estimates for derivatives of FBI terms. At the end of this section we will present a general inequality for the FBI transform acting on ultradistributions.
Lemma 3.1
Let \(\omega \) be a weight function (see Definition 2.1) and \(\kappa >0\). If \(\mu _{\omega }\) and \(\delta =\delta _{\omega }\) are defined as in Proposition 2.20 then, there exist \(\delta ,D>0\) such that,
for all \(\xi \in \mathbb {R}^n\) \(x,y \in \mathbb {R}\), \(m \in \mathbb {N}_0\) and \(\theta >0.\)
Proof
Note that, using Faà di Bruno’s formula (see [14, Corollary 2.11]) it follows that
where \({\mathfrak {p}}(m, r)= \left\{ (k_1, \dots , k_m) \in {\mathbb {N}}_0^m: \ \sum _{j=1}^mk_j=r, \ \sum _{j=1}^m j k_j=m \right\} .\)
Since \(\partial _{y}^j \{-\kappa \mu _{\omega }(\xi )(x-y)^2\}=0,\) for \(j\ge 3,\) we will be able to consider the sum over a subset of \(\mathfrak { p }(q,r)\), considering only derivatives of order less than three. Hence, denoting \({\mathfrak {p}}_2(m, r)= \left\{ (k_1, k_2) \in {\mathbb {N}}_0 ^2: \ k_1+k_2=r, \ k_1+2k_2=m \right\} ,\) we have
And, using that \(r=k_1+k_2 =\frac{k_1}{2}+\frac{m}{2}\), \({k_1} \le {m}\), \(m! \le 2^{m} r! k_2!\) and \(k_2!k_1!^{1/2}=(k_2!^2k_1!)^{1/2} \le m!^{1/2}\) we obtain,
where \(D>1\) is as in (2.12). Recalling that (see [14, p. 515]),
and
using (2.11) and (2.12) it follows that
Thus, for each \(\theta >0\) it follows that,
\(\square \)
Remark 3.2
Observe that \(\partial _{y}\left\{ e^{-\kappa \mu _{\omega }(\xi )(x-y)^2} \right\} = -\partial _{x}\left\{ e^{-\kappa \mu _{\omega }(\xi )(x-y)^2} \right\} , \) for \(\xi \in \mathbb {R}^n\) and \(y,x \in \mathbb {R}.\) Thus, we can rewrite (3.1) by
for all \(\xi \in \mathbb {R}^n\), \(x,y \in \mathbb {R}\), \(\ell ,m \in \mathbb {N}_0\) and \(\theta >0.\)
Remark 3.3
Observe that
for each \(x,\xi ,y=(y_1,\dots ,y_n)\in \mathbb {R}^n\) and \(\alpha =(\alpha _1,\dots ,\alpha _n)\in \mathbb {N}_0.\) Thus, it follows from Lemma 3.1,
Next we will present bounds on the derivatives of the FBI phase.
Lemma 3.4
If \(\omega \) is a weight function, \(\mu _{\omega }\) and \(\delta =\delta _{\omega }\) are defined as in Proposition 2.20, \(\kappa >0\) and \(Q_{\omega }^{\kappa }(z,y,\xi )=i(x-y)\cdot \xi -\kappa \mu _{\omega }(\xi ) (x-y)^2,\) then for each \(\theta ,\lambda >0\) there exists \(D=D_{\theta ,\lambda }\) such that
for \((\gamma ,\alpha ,x,y,\xi ) \in \mathbb {N}_0^n\times \mathbb {N}_0^n \times \mathbb {R}^n \times \mathbb {R}^n \times \mathbb {R}^n\).
Proof
From Leibniz rule it follows that,
where \(\beta =(\beta ',\beta ''),\) \(\beta '=(\beta '_1, \dots ,\beta '_n)\in \mathbb {N}_0^n\) and \(\beta ''=(\beta ''_1, \dots ,\beta ''_n)\in \mathbb {N}_0^n\).
Therefore, using (3.3), (2.3), (2.7) and (2.8) there exists \(C,D>1\) such that,
for each \(\lambda ,\theta >0.\) \(\square \)
The following result consists of a general domination of the FBI transform for ultradistributions.
Lemma 3.5
Let \(\Omega \subset \mathbb {R}^n\) be an open subset. If \(\omega \) is a weight function, \(\mu _{\omega }\) and \(\delta =\delta _{\omega }\) are defined as in Proposition 2.20, \(\kappa >0,\) \(u \in {\mathcal {D}^{\omega }}'(\Omega )\) and \(\phi \in {\mathcal {D}^{\omega }}(\Omega ) \) with \(K=\text{ supp }\,\phi \). Then, there exists \(\lambda >0\) such that for each \(\theta >0\) we can find \(C=C_{\theta }>0\) such that
for each \((x,\xi ) \in \mathbb {R}^n \times \mathbb {R}^n\) and \(\gamma \in \mathbb {N}_0^n.\)
Proof
Observe that denoting \(e_1=(1,0, \dots ,0), \dots , e_n=(0, \dots , 0,1)\) and considering a real sequence \(\{h_j\}_{j=1}^{\infty }\) such that \(h_j \rightarrow 0\) it follows that, if \(\psi \in \mathcal {E}^{\omega }(\mathbb {R}^n \times \mathbb {R}^n)\) is such that \(\psi (x,y)=\psi _1(y)\psi _2(x,y)\) (where \(\psi _1\in \mathcal {D}^{\omega }( \mathbb {R}^n)\) and \(\psi _2 \in \mathcal {E}^{\omega }(\mathbb {R}^n \times \mathbb {R}^n)\)), then
in \(\mathcal {D}^{\omega },\) for each \(k \in \{1, \dots n\}\) and \(x \in \mathbb {R}^n\). Thus, using the continuity of u
Hence, it follows that
where \(\gamma \in \mathbb {N}_0^n.\) Moreover, from Lemma 2.17 for each \(\epsilon >0\) there exists \(C_{\epsilon }>0\) such that
Since \(\phi \in \mathcal {D}^{\omega },\) there exists \(\lambda _1>0 \) such that \(\Vert \partial ^{\beta ''}\phi \Vert _{\infty } \le e^{\frac{1}{2\lambda _1} \varphi ^*(2\lambda _1 |\beta ''|)},\) for each \(\beta '' \in \mathbb {N}_0^n.\) Moreover, using Leibniz rule, (2.3), Lemma 3.4 and Proposition 2.20, for each \(\lambda ,\theta >0\) there exist \(D>0\) such that
where \(\lambda _2=\max \{\lambda ,\lambda _1\}\). Next we will consider \( \epsilon = \frac{1}{\lambda _2 L^k}\), where \(k \in \mathbb {N}\) is chosen such that \(e^{-k} (1+D)<1\) and L is as in (\(\alpha \)). From, (2.4),
for some \(C'>0.\) From the arbitrariness of \(\lambda \) and \(\theta \) the proof is completed. \(\square \)
4 Proof Theorem 1.1: Necessity
The aim of this section is to prove the necessary condition of Theorem 1.1 i.e, when given an ultradistribution u such that u is \(\mathcal {E}^{\omega }\) regular in a neighborhood of a given point \(x_0\), then (1.5) is satisfied. To do so we will prove a more general result, Theorem 4.2, in which the necessity part of Theorem 1.1 will be a particular case (just choose \(\alpha =0\) and \(\kappa =1\) in Theorem 4.2).
The reasons to present this more general result are the following. First, we need the result for arbitrary \(\alpha \ne 0\) for the FBI inversion formula for ultradifferentiable functions presented in Sect. 5.1, see Lemma 5.2 and its consequences, Lemma 5.3 and Lemma 5.5.
Second, we deal with any parameter \(\kappa > 0\) since in the proof of Lemma 5.5 we consider a general \(\kappa \) and use Lemma 5.3 with \(\kappa /2\) (see (5.14)). For this reason it is required to allow arbitrary \(\kappa > 0.\) Moreover, Lemma 5.5 is important in the proof of the sufficiency part of Theorem 1.1.
Following the main ideas of [18] we start by presenting a characterization of ultradistributions vanishing in a neighborhood of a given point. And then we will use this fact to prove the main result of this section.
Lemma 4.1
Let \(\sigma \) be a weight function, \(\kappa >0,\) \(x_0 \in {\mathbb {R}}^n, \) \(\Omega \subset \mathbb {R}^n\) be an open neighborhood of \(x_0\) and \(u \in {\mathcal {D}^{\sigma }}'(\Omega )\). If u vanishes in a neighborhood of \(x_0,\) then for each \(\phi \in \mathcal {D}^{\sigma }(\Omega )\) there exist constants \(C,c,\lambda >0\) and a neighborhood \(V\subset \Omega \) of \(x_0\) such that
Proof
Given \(\phi \in {{{\mathcal {D}}}^{\sigma }}(\Omega )\) it follows that \(\phi \in {{{\mathcal {D}}}^{\sigma }}(\mathbb {R}^n)\) and there exists \(R>0\) such that \(\text{ supp } \,\phi \subset B(x_0,R) \). Considering \(u\in {{{\mathcal {D}}}^{\sigma }}'(\Omega )\) vanishing in a neighborhood of \(x_0\), we can assume that there exists \(0<r <R/3\) such that \(u \equiv 0\) in \(B(x_0,3r)\). Also, consider a function \(\psi \in \mathcal {D}^{\sigma }(B(x_0,3r))\) such that \(\psi \equiv 1\) in \(B(x_0,2r).\) Then, \({supp} \{\phi \psi \} \subset B(x_0,3r) \) and \({supp} \{\phi (1-\psi )\} \subset B(x_0,R) {\setminus } \overline{B(x_0,r)}.\) Thus, using Lemma 3.5, there exists \(\lambda >0\) such that for each \(\theta >0\) there exists \(C=C_{\theta }>0\) such that
for every \((x,\xi ,\alpha ) \in \mathbb {R}^n \times \mathbb {R}^n \times \mathbb {N}_0^n\) where \(a= \frac{\kappa }{2}\). Moreover, if \(\Vert x-x_0\Vert < r/2\) and \(\Vert y-x_0\Vert > r\) then \(\Vert x-y\Vert > r/2.\) Thus, since \(\delta >0\) (here \(\delta \) denotes the constant \(\delta _{\sigma }\) as defined in Proposition 2.20.), recalling that \(\sigma \) is an increasing function and considering \(\theta =\frac{ar^2}{8}\) we obtain
Thus setting \(V=B(x_0,{r/2})\) finishes the proof. \(\square \)
Next we shall present the main result of this section whose consequence is the necessity part of Theorem 1.1. Note that if \(\sigma \) and \(\omega \) are weight functions with \(\omega (t)=O(\sigma (t))\) as \(t \rightarrow + \infty \) then \(\mathcal {D}^{\sigma }(\Omega )\subset \mathcal {D}^{\omega }(\Omega )\) (see Remark 2.10) and \({\mathcal {D}^{\omega }}' (\Omega )\subset {\mathcal {D}^{\sigma }}' (\Omega )\).
Theorem 4.2
Let \(\omega \) and \(\sigma \) be weight functions such that \(\omega (t)=O(\sigma (t))\) as \(t \rightarrow + \infty \), \(\kappa >0\), \(\Omega \subset {\mathbb {R}}^n\) be an open set and \(u \in {\mathcal {D}^{\sigma }}' (\Omega )\). If u is \(\mathcal {E}^{\omega }\) in a neighborhood of \(x_0\in \Omega \) then for each \( \phi \in \mathcal {D}^{\sigma }(\Omega )\) there exist \(C,c, \lambda >0\) and a neighborhood \( V\subset \Omega \) of \(x_0\) such that
where \(\varphi ^*=\varphi _{\omega }^*\) is the Young conjugate of \(\varphi (t)=\omega (e^t).\)
Proof
Fix \(R>0\) such that \(u\in \mathcal {E}^{\omega }(B(x_0,R))\) and \(B(x_0,R) \subset \Omega \). Consider \(\psi \in \mathcal {D}^{\sigma }(B(x_0,R))\) such that \(\psi \equiv 1\) in \(B(x_0, \tfrac{R}{2})\) and write
Now we use Lemma 4.1 to obtain constants \(C,a,\lambda >0\) and a bounded neighborhood V of \(x_0\) such that
Furthermore, since \(\omega (t)=O(\sigma (t))\) as \(t\rightarrow + \infty \) there exist \(A,c>0\) so that \(\omega (\xi )\le c\sigma (\xi )\) for \(\Vert \xi \Vert >A\) and by the continuity of \(\omega \), for each \(c>0\) there exists \(C_c>0\) such that \(1= e^{c\omega (\xi )} e^{-c\omega (\xi )} \le C_c e^{-c\omega (\xi )}\) for \(\Vert \xi \Vert \le A\). Thus, we use Remark 2.10, define \(a_1= \frac{a}{c}\), increase \(\lambda , C>0,\) if necessary, and obtain
Next we will fix \(\xi =(\xi _1, \dots , \xi _n) \in \mathbb {R}^n{\setminus } \{0\}\) and consider \(j\in \{1, \dots , n\}\) such that \(\Vert \xi \Vert \le n |\xi _j|\). Moreover, given an arbitrary \(N \in \mathbb {N}_0 \), integrating by parts with respect to \(y_j\), we can write
for \(x,\xi \in \mathbb {R}^n,\) where it was used that \(\psi \phi u \in \mathcal {D}^{\omega }(B(x_0,R))\subset \mathcal {D}^{\omega }(\mathbb {R}^n)\). Given \(\alpha \in \mathbb {N}_0^n\), we have
Now, Remark 2.8, (2.5), (2.8) together with Leibniz’s rule and Proposition 2.20 (where it is proved that the derivatives of \(a_{\mu _{\sigma }}\) are uniformly bounded in compact sets) show that there exist \(C_2,\lambda _2>0\) (independent of x and \(\xi \)) such that
Furthermore, it follows from inequality (3.2) that there exist \(D>0\) such that
Denote \(\xi _{\sigma }\doteq {\sigma }( ({\Vert \xi \Vert ^2 + \delta _{\sigma }})^{1/2})\), increase D (if necessary) and use (2.7) to obtain
Using Remark 2.11 we can find \(D_1>D\) such that \(\omega (t) \le t,\) for each \(t>D_1^2\) and since \(\omega \) is increasing we can write
Thus, using Leibniz’s rule, Remark 2.8, (4.6) and (4.9) there exists \(m=m(n)>0 \) such that
where we considered \(\lambda _1=\lambda _2\), used the notation \(\beta =(\beta _1,\beta _2), \gamma =(\gamma _1,\gamma _2) \in \mathbb {N}_0^n \times \mathbb {N}_0^n\) and denoted \(C_{\lambda }=m+me^{\frac{D_1^2}{2\lambda }} \) (for \(\lambda >0\)).
Hence, recalling that \(\Vert \xi \Vert \le n|\xi _j|\), denoting \({\tilde{C}}=C_2 \,|\text{ supp }\, \psi \, \phi |\) (where \(\,|\text{ supp }\, \psi \, \phi |=\int _{\text{ supp }\, (\psi \, \phi )}1\,dy\)) and using (4.10) one can estimate the expression in (4.5) by
Denoting \(a=\frac{1}{nC_{\lambda _1}}\) and using Remark 2.8, (2.7) and (2.3) there exists \(k_{\lambda _0}=k(\lambda _,a)>0\) such that
Taking the infimum in \(N \in \mathbb {N}_0\) in the last inequality we obtain
Recalling that \(\inf _{N \in \mathbb {N}_0} t^{-N} e^{\frac{1}{\lambda _1} \varphi ^*({\lambda _1}N)} \le e^{\log t}e^{-\frac{1}{\lambda _1}\omega (t)}\) for each \(t\ge 1\), see [12, page 218],then by increasing \({\tilde{C}}\) if necessary, it follows that one can further estimate the expression in (4.12) by
Now, we take advantage of \(\gamma \)) and Remark 2.11 to obtain \(A_1>0\) such that
\(\xi _{\sigma }=\sigma (({ \Vert \xi \Vert ^2+\delta _\sigma })^{1/2}) \le {a}\Vert \xi \Vert \) and \(\log (a\Vert \xi \Vert ) \le \frac{1}{{4}\lambda _1}\omega (a\xi ),\) when \(\Vert \xi \Vert >A_1\).
Thus, choosing \(\lambda _0=12\lambda _1\)
for each \(x \in V\) and \(\Vert \xi \Vert > A_1.\) Now, we choose \(k \in \mathbb {N}\) satisfying \(e^{-k}\le a\). Hence, using (2.2) we see that there exists \(A_2\ge A_1\) such that
Thus, using (2.5) there exist \(a_1>0\) and \(\lambda _3>0\) such that one can further estimate (4.13) by
This concludes the proof when \(\Vert \xi \Vert >A_2\). For \(\xi \) small, we go back to (4.5) and use (4.10) to get (possibly by increasing m)
Since \(\omega \) and \(\sigma \) are continuous function (consequently \(\sup _{\Vert \xi \Vert \le A_2} e^{\frac{1}{2\lambda _1} \omega (\xi _\sigma )}<+\infty \)), using (2.5), there exists \(\lambda _2, \,{\tilde{C}}_1, \, {\tilde{C}}_2>0\) such that
for \(\Vert \xi \Vert \le A_2,\) \({\tilde{C}}_2 \ge {\tilde{C}}_1 \sup _{\Vert \xi \Vert \le A_2} e^{a_1\omega (\xi )}.\) Therefore, it follows from (4.3), (4.4), (4.14) and (4.15) that the proof is completed. \(\square \)
An immediate consequence is the necessity of Theorem 1.1.
Corollary 4.3
Fix a weight function \(\omega \), an open set \(\Omega \subset {\mathbb {R}}^n\) and \(u\in {\mathcal {D}^{\omega }}' (\Omega )\). If \(u \in \mathcal {E}^{\omega }\) in a neighborhood of \(x_0\in \Omega \) then there exist a weight function \(\sigma \) such that \(\omega (t)=O(\sigma (t))\) as \(t \rightarrow + \infty \), so that for each \(\phi \in \mathcal {D}^{\sigma }(\Omega )\) there exist \(C,c>0\) and a neighborhood \(x_0\in V\subset \Omega \) such that
Proof
Under the corollary hypothesis one can select any weight function \(\omega (t)=O(\sigma (t))\) as \(t \rightarrow + \infty \) then the proof of Theorem 4.2 will work for \(u\in {\mathcal {D}^{\omega }}' (\Omega )\subset {\mathcal {D}^{\sigma }}' (\Omega )\). \(\square \)
5 Proof Theorem 1.1: Sufficiency
In order to prove the sufficiency part of Theorem 1.1 we shall prove an inversion formula for the FBI transform \(\mathcal {F}^{\kappa }_{\mu _{\omega }}\) defined in (1.4).
5.1 Inversion Formula for Ultradifferentiable Functions
Throughout this section we will consider a weight function \(\omega \) and we will denote by \(\mu _{\omega }\) the function defined in Proposition 2.20. In order to prove the sufficient condition of Theorem 1.1 in this section we will present two inversion formulas of \({{\mathcal {F}}}^{\kappa }_{\mu _{\omega }}u\) when u is in \(\mathcal E^{\omega }\), \(\kappa >0\) and \(\omega \) is any weight function.
Lemma 5.1
If \(u \in C_c^{n+1}({\mathbb {R}}^n)\) then
Proof
Let \(R>1\) be such that \(\text{ supp }\,u \subset B(0,R).\) Since \(u \in C_c^{n+1}({\mathbb {R}}^n)\) it follows that \(\sup _{\xi \in {\mathbb {R}}^n} \{(1+\Vert \xi \Vert )^{n+1} |{\hat{u}}(\xi )| \}< + \infty \). Thus, using the Fourier inversion formula we can write
Define \(\Gamma = \Gamma (x,x', \xi , t) \doteq \xi +i\, t \, \kappa \, \mu _{\omega }(\xi )\, (x-x'), \) for \(x,x',\xi \in \mathbb {R}^n\) and \( t \in \mathbb {R}\). Since \(\mu _{\omega } \ge 0\),
This, together with the fact that the function \(\omega \) is increasing and (2.12) we see that there exist constants \(D,\delta >0\) such that
for an arbitrary \(r>0.\)
Moreover, using (2.6) there exists a positive constant \({\tilde{A}}\) such that one can further estimate the last expression as follows
Since \(\xi \mapsto e^{- \epsilon {\tilde{a}} \Vert \xi \Vert ^2} \) is a \(L^1\) function for each \(\epsilon >0\), using (5.2) it follows that
Now, using that \(\zeta \mapsto u(x')e^{i(x-x') \cdot \zeta -\epsilon \zeta ^2 } \) is holomorphic, we may apply Stokes’ theorem together with the definition of \(a_{\mu _{\omega }}\) (see (1.3)) to obtain,
Thus, as a consequence of the third item in Proposition 2.20, inequality (5.2) and Fubini’s theorem, we can further express (5.3) as
Now, we will show that we can use the dominated convergence theorem in the right-hand side of (5.4).
The goal is to bound the integral in \(x'\) uniformly in \(\epsilon \) by an \(L^1\) function in \(\xi \). Let \(Q_{\kappa }(x,x',\xi )\doteq -\epsilon (\xi +i\kappa \mu _{\omega }(\xi )(x-x'))^2\) and note that it follows from (2.6) that there exist constants \({\tilde{A}},{a}>0\) so that
Therefore, the trivial bound is not good enough for our purpose. The trick here is to use integration by parts and to do so fix \(\xi \in {\mathbb {R}}^n \) and assume without loss of generality that \(|\xi _1|=\max \{|\xi _k|: k \in \{1, \dots , n\}\}.\) Then it follows that
for \(x' \in B(0,R)\) and \(x \in B(0,r)\). Moreover, using (2.6) and (2.12) and increasing \({\tilde{A}}>0\) (if necessary) we have
where D is the constant appearing in (2.12). Thus one can use (5.7) to continue estimating (5.6) as
Thus, integrating by parts the integral in \(x'\) in the right-hand side of (5.4) multiplied by \(\Vert \xi \Vert ^{n+1}\), we have
for \(x' \in B(0,R)\), \(x \in B(0,r)\) and \({\Vert \xi \Vert }>{\tilde{A}}.\)
Moreover, for each \(q \in \{1,\dots , n+1\}\), using Faà di Bruno’s formula (see [14, Corollary 2.11])
and
It is easy to see that there exists a constant \(c>0\) such that
\( |\partial _{x'_1}Q_{\kappa }(x,x',\xi )| \le \epsilon c \Vert \xi \Vert ^2, \quad |\partial _{x'_1}^2Q_{\kappa }(x,x',\xi )| \le \epsilon c \Vert \xi \Vert ^2, \quad \text{ and } \quad \partial _{x'_1}^pQ_{\kappa }(x,x',\xi ) =0 \ (p \ge 3),\)
for \(x' \in B(0,R)\), \(x \in B(0,r)\) and \(\Vert \xi \Vert >{\tilde{A}}\).
Consequently, we will be able to consider the sum over a subset of \({\mathfrak {p}}(q,r)\), considering only derivatives of order less than three. Moreover, since \(\displaystyle {r!\sum \nolimits _{\mathfrak { p }(q,r)} \prod \nolimits _{p=1}^q \frac{1}{k_p!}=\left( {\begin{array}{c}q-1\\ r-1\end{array}}\right) }\) (see [14, p. 515]), it follows that
where \({{\mathfrak {p}}_2(q,r)}=\{(k_1,k_2): k_1+k_2=r \,\,\, \text{ and } \,\,\, k_1+2k_2=q\}\), for \(x' \in B(0,R)\), \(x \in B(0,r)\) and \(\Vert \xi \Vert >{\tilde{A}}\). Thus using the Leibniz rule and (5.10) one can bound the term in (5.9) by an uniform constant \(C_n>0\) independent of \(\epsilon \),
Since u is compactly supported, considering (5.4) and taking into account (5.11) we can use the dominated convergence theorem in the right-hand side of (5.4) to conclude that
for a arbitrary \(r>0\) as we wished to prove. \(\square \)
Lemma 5.2
If \(\kappa >0\) and \(u \in \mathcal {D}^{\omega }({\mathbb {R}}^n)\) then, for \( x \in {\mathbb {R}}^n:\)
Proof
Observe that, from Theorem 4.2 for each compact subset \( K \subset \mathbb {R}^n\) there exist an open neighborhood V of K and \(a,\lambda ,C>0\) such that
Moreover, using Lemma 5.1 it follows that \(u(x)=(2\pi )^{-n} \int _{\mathbb {R}^n} \mathcal {F}^{\kappa }_{\mu _{\omega }} u (x,\xi ) d \xi .\) Thus, from (5.12),
The lemma now follows from (2.9) and the dominated convergence theorem. \(\square \)
5.2 FBI Inversion Formulas for Ultradistributions
The aim of this section is to present two inversion formulas for ultradistributions.
Lemma 5.3
Let \(\kappa >0\) be arbitrary. If \(u \in {{{\mathcal {D}}}^{\omega }}'({\mathbb {R}}^n)\) and \(\psi \in {{{\mathcal {D}}}^{\omega }}({\mathbb {R}}^n)\) then
where \(u_{j}(x) \doteq (2\pi )^{-n} \int e^{-\frac{\Vert \xi \Vert ^2}{j}} {{\mathcal {F}}}_{\mu _{\omega }}^{\kappa } (\psi u) (x,\xi ) d\xi .\)
Remark 5.4
Note that from Lemma 3.5 and Remark 2.11 it follows that \(u_j\) is well defined.
Proof
For each \(\phi \in \mathcal {D}^{\omega }({\mathbb {R}}^n)\) using Lemma 5.2 and the notation \({\check{\phi }}(x)=\phi (-x)\) we have
Where in the second equality we can apply similar arguments as in the proof of Lemma 3.4 to obtain the convergence of the Riemann integral in \(\mathcal {D}^{\omega }-\)topology.
Also, in the second and third equality, it was used that \(e^{-\frac{{\Vert \cdot \Vert }^2}{j}} \in L^1 (\mathbb {R}^n)\) and that the support of \(\psi \) and \(\phi \) are compact subsets of \({\mathbb {R}}^n\). \(\square \)
Next we will use Lemma 5.3 to obtain another inversion formula which will be used in the proof of Theorem 5.6.
Lemma 5.5
Let \(\kappa >0\) be arbitrary. If \(u \in {{{\mathcal {D}}}^{\omega }}'({\mathbb {R}}^n)\) , \(\psi \in {{{\mathcal {D}}}^{\omega }}({\mathbb {R}}^n)\) then
where \({\tilde{u}}_{j}(x) \doteq (2\pi ^3)^{-n/2} \int \int e^{i\xi (x-t)-\kappa \mu _{\omega } (\xi )(x-t)^2} e^{-\frac{\Vert \xi \Vert ^2}{j}} {{\mathcal {F}}}^{\kappa }_{\mu _{\omega }} (u\psi ) (t,\xi ) \,\,\big (\kappa \mu _{\omega }(\xi )\big )^{n/2}dt d\xi .\)
Proof
Since, \(e^{-\mu _{\omega }(\xi )(x-\cdot )^2} \in L^1(\mathbb {R}^n)\) (for each fixed \(x,\xi \in \mathbb {R}^n\)) we can rewrite \({\tilde{u}}_{j}(x)\) as
where it was used that the Riemann sum converges to the Riemann integral in the \(\mathcal {D}^{\omega }\)-topology with respect to the variable t. Using the equations
and
it follows that
Since \( e^{-2\kappa \mu _{\omega }(\xi ) (\cdot -\frac{1}{2}(x+x'))^2 } \in L^{1}(\mathbb {R}^n)\) (for each \(\xi ,x,x',\kappa \)) we consider the change of variables, \(-w=t-\frac{1}{2}\left( x+x' \right) \) to write
Moreover, using the identity (see [32, (IX.4.6)])
Observe that \(a_{\mu _{\omega }}^{\kappa } \big (z,\xi \big )\) is a polynomial of degree n with respect to \(z\in \mathbb {R}^n\) (see (1.3)). Additionally, it is a polynomial of degree 1 as a function of \(z_j\) for each \(j\in \{1,\dots ,n\}\), depending on the parameters \(z_k\) for \(k\ne j\). Since \(\kappa >0\) and \(\mu _{\omega }(y) >0 \) we obtain
This together with (5.13) allow us to rewrite \({\tilde{u}}_{j}(x)\) as
Where in the second equality we used that \(a^{\kappa }_{\mu _{\omega }}\left( \frac{x}{2},\xi \right) =a^{{\kappa }/{2}}_{\mu _{\omega }}\left( x,\xi \right) \). Therefore the result follows from Lemma 5.3. \(\square \)
5.3 Sufficient Condition of Theorem 1.1
In this section we will use the inversion formula presented in the previous section to prove that a certain decay of the FBI transform in all directions implies \(\mathcal {E}^{\omega }\) local regularity.
Theorem 5.6
Let \(x_0 \in \mathbb {R}^n,\) \(\omega \) and \(\sigma \) be weight functions such that \(\omega (t)=O(\sigma (t))\) (for \(t \rightarrow + \infty \)), and \(u \in {\mathcal {D}^{\sigma }}' (\mathbb {R}^n)\). If there exist \( \psi \in \mathcal {D}^{\sigma }(\mathbb {R}^n)\) and \(C,c,\kappa ,r>0\) such that
then \(u \in \mathcal {E}^{\omega }\) in a neighborhood of \(x_0\).
Proof
We first note that it follows from Remark 2.11 and inequality (2.12) that there exists \(A_1>1\) such that \(|\sigma (( \Vert \xi \Vert ^2+\delta _\sigma )^{1/2})| \le e^{-c}\Vert \xi \Vert \) and \(|\mu _{\sigma }(\xi )|\le \Vert \xi \Vert ,\) for \({\Vert \xi \Vert }>A_1.\) Moreover, using (2.7) we obtain
Moving on we want to apply Lemma 5.5 and in order to do so we write
and
where, \(U_1=\{(t,\xi ): \, \Vert t-x_0\Vert <r, \,\, \Vert \xi \Vert >A_1 \},\) \(U_2=\{(t,\xi ): \, \Vert t-x_0\Vert <r, \,\, \Vert \xi \Vert \le A_1 \},\) \(U_3=\{(t,\xi ): \, \Vert t-x_0\Vert \ge r, \,\, \xi \in \mathbb {R}^n\}\). Next we will prove that for each \(\ell \in \{1,2,3\}\) there exists a function \(I_\ell \in \mathcal {E}^{\omega }\) in a neighborhood \(U_0\) of \(x_0\) such that \(\displaystyle \lim _{j \rightarrow + \infty }\int _{U_\ell }\Psi (x,t,\xi ,j)dtd\xi = I_\ell (x)\) in \(\mathcal {E}^{\omega }(U_0)\). Therefore, using Lemma 5.5 we will obtain \(u=I_1+I_2+I_3\) in a neighborhood of \(x_0,\) concluding the proof.
Observe that \(\int _{\Vert t-x_0\Vert<r} \int _{\Vert \xi \Vert >A_1} e^{-c\omega (\xi )} d\xi dt <+ \infty .\) Thus, we can use the dominated convergence theorem to conclude that
as \(j\rightarrow \infty .\) Moreover, using (3.2) and (5.15) there exists \(D>0\) such that, for every \(x \in \mathbb {R}^n\) and every \(\alpha \in \mathbb {N}_0^n\) it holds
Since \(|\sigma (({ \Vert \xi \Vert +\delta _\sigma })^{1/2})| \le e^{-c}\Vert \xi \Vert \) and \(|\mu _{\sigma }(\xi )|\le \Vert \xi \Vert ,\) for \(\Vert \xi \Vert >A_1,\) using (2.3), (2.7) and (2.8) there exists \(D_0>0\) such that
Thus, using (2.7) there exists \(C_1,D_1>0\) such that
Since \(t \mapsto \frac{\varphi ^*(t)}{t}\) is increasing, using (2.3), the last inequality can be further estimated as follows
Now we invoke (2.5) to obtain \(D_2>0\) and \(\lambda _1>0\) such that \(|\partial _x^{\alpha }I_1(x) | \le D_2 e^{\frac{1}{\lambda _1} \varphi ^*(\lambda _1 |\alpha |)},\) for each \((x,\alpha )\in \mathbb {R}^n \times \mathbb {N}_0^n.\) Thus \(I_1 \in \mathcal {E}^{\omega }({\mathbb {R}}^n).\) Moreover, reasoning analogously we see that,
Therefore,
Next we will study \(U_2.\) Since \(U_2\) is bounded for each \(x \in \mathbb {R}^n\) it follows that
Also using (3.2) and (5.15) there exist constants \(C,D>0\) such that for every \(\alpha \in \mathbb {N}_0^n\) we have
Hence, there exist \(D_3,D_4>0\) such that \(\Vert \partial _x^{\alpha }I_2 \Vert \le D_3 D_4^{|\alpha |} |\alpha |!\), for every \(\alpha \in \mathbb {N}_0^n.\) Therefore, using Remark 2.12 it follows that \(I_2 \in \mathcal {E}^{\omega }({\mathbb {R}}^n)\). Moreover, one can see that
Thus,
Next, in order to study the integral in the region \(U_3\), observe that
for \(\Vert t-x_0\Vert \ge r\) and \(\Vert x-x_0\Vert <\frac{r}{2}\).
Thus, using (2.11), (2.12) and Lemma 3.5 there exists \(D>0\) such that,
for \(\tilde{\sigma }(\xi )=\sigma ( (\delta _{\sigma }+\Vert \xi \Vert ^2)^{1/2})\), \((t,\xi ) \in U_3\) and \(\Vert x-x_0\Vert <\frac{r}{2},\) where it was used that \(\left[ \frac{(\kappa r^2 \tilde{\sigma }(\xi ))^n}{16^n n!} \right] ^{\frac{1}{2}} \le \left[ e^{\frac{\kappa r^2 \tilde{\sigma }(\xi )}{16}} \right] ^{\frac{1}{2}}\). Hence, using the dominated convergence theorem (recalling that \( \int e^{-\kappa \frac{r^2}{16} \tilde{\sigma }(\xi )}d\xi < + \infty \) and \(\int e^{-\kappa \sigma (\delta ^{1/2})(x-t)^2} dt < + \infty \)),
Moreover, using the Leibniz rule, Remark 3.3, Lemma 3.5, (2.3), (2.7) and (2.8), for each \(\theta ,\lambda >0\) to be chosen there exist \(D_3,D_4,D_5>0\) such that
where in the last inequality we fixed \(\theta \doteq \frac{\kappa r^2 }{ 32}\) and \(\kappa _{\sigma } \doteq \frac{\kappa }{16} \sigma (\sqrt{\delta })\). Moving on using (2.5) we see that there exist \(D_6>0\) and \(\lambda _*>0\) such that the last inequality can be further estimated as
Moreover, from the fact that
and considering \(A_3,c_1>0\) so that \(\omega (\xi ) \le c_1 \sigma (\xi ),\) for \(\Vert \xi \Vert >A_3,\) it follows that
Note that, the first integral is finite and choosing \(\lambda \doteq \frac{{64}\, c_1}{\kappa r^2 }\) it follows from (2.12), (2.7) and Remark 2.11, that
Thus, summing up (5.19), (5.20) and (5.21) we see that there exists \(D_7,\lambda _*>0\) such that
This shows that \(I_3 \in \mathcal {E}^{\omega }.\) In addition to that and similarly as before it follows that
Hence
Therefore, using (5.17), (5.18), (5.22) and Lemma 5.5 it follows that \(u=I_1+I_2+I_3\) in \(B(x_0,\tfrac{r}{2}),\) which concludes the proof. \(\square \)
6 A Characterization of Ultradifferentiable Iterates of Constant Coefficients Operators
Let \(P(\xi )= \sum _{|\alpha | \le m} a_{\alpha } \xi ^{\alpha }\) be a polynomial function of degree m. This section is dedicated to the characterization of the space \(\mathcal {E}^{\omega }(\Omega ;P)\) (see Definition 1.2) using a FBI transform.
Remark 6.1
From now on we will consider \(0<\rho \le 1\) such that (1.8) is satisfied and denote \(\sigma (t)=\sigma _{\omega ,\rho }(t)=\omega (t^{\rho }).\) It is important to note that, if \(\omega \) is a weight function and \(0<\rho \le 1\) then \(\sigma \) is a weight function.
Theorem 6.2
Let \(x_0 \in \mathbb {R}^n,\) \(\omega \) be a weight function and P(D) be a constant coefficient hypoelliptic linear operator of order m together with its hypoelliptic index \(\rho \) (satisfying (1.8)). Let \(\sigma (t)=\omega (t^{\rho })\) and \(u \in {\mathcal {D}^{\sigma }}' (\mathbb {R}^n)\subset {\mathcal {D}^{\omega }}' (\mathbb {R}^n)\). The following conditions are equivalent:
-
1.
There exists a neighborhood \(\Omega \) of \(x_0\) such that \(u \in \mathcal {E}^{\omega }(\Omega ;P)\).
-
2.
There exist \(\phi \in \mathcal {D}^{\omega }(\mathbb {R}^n)\) (such that \(\phi \equiv 1\) in a neighborhood of \(x_ 0\)), \(C,\lambda , c>0\) and a neighborhood V of \(x_0\) such that
$$\begin{aligned} |\mathcal {F}_{\mu _{\sigma }} \left( \phi \, [P(D)]^{N} ( u)\right) (x,\xi ) | \le C e^{\frac{1}{\lambda } \varphi ^*(Nm \lambda )} e^{-c \omega (\Vert \xi \Vert ^{\rho })}, \quad (x,\xi ,N) \in V \times \mathbb {R}^n \times \mathbb {N}_0,\nonumber \\ \end{aligned}$$(6.1)
where we are denoting \( \varphi ^*(x)=\varphi _{\omega }^*(x) \doteq \sup \{xy-\varphi (y): \, y \ge 0\},\) for each \(x>0\) and \(\mu _{\sigma }\) is the function obtained from Proposition 2.20 when applied to \(\sigma \) instead of \(\omega .\)
Proof
\({ (1) \Rightarrow (2) }\): Let \(R\in (0,1/3)\) be such that \(B(x_0,3R) \subset \Omega \). Set \(0<r<R\) and \(\phi \in \mathcal {D}^{\omega }(B(x_0,R))\) such that \(\phi \equiv 1\) in \(B(x_0,r).\) Also, considering \(\psi \in \mathcal {D}^{\omega } (B(x_0,2R))\) such that \(\psi \equiv 1\) in \(B(x_0,R),\) it follows that
where \(x,\xi \in \mathbb {R}^n\),
and
Next we will study \(I_1\) and \(I_2.\) First, denoting \(K=\text{ supp } \psi \) and using Hölder inequality,
Since \(u\in \mathcal {E}^{\omega }(\Omega ;P) \) and \(K \subset \Omega \), recalling Definition 1.2, there exists \(C_1,C>0\) and \(\lambda >0\) such that
for each x satisfying \(|x-x_0|\le \frac{r}{2}\) and \(\xi \in \mathbb {R}^n;\) where we also use Proposition 2.20.
Next we will consider the term \(I_2\) given by (6.3). Let \(\ell \) be an arbitrary positive integer. Using (1.8) we obtain for \(\Vert \xi \Vert \ge K,\) with K the constant appearing in (1.8), and \(u_n:= [P(D)]^Nu\)
In order to study the above integral we first recall that
where we denote \(P^{(\beta )}g \doteq (\partial _{\xi }^{\beta }P)(D)g\) for each \(\beta \in {\mathbb {N}}^n_0\). Using the linearity of P(D) it follows that
Hence,
Thus, inserting the last expression in (6.4) it follows that,
where \(\Vert \xi \Vert >K\). Observe that, since \(\sigma \) is increasing it follows that \(\sigma ((\Vert \xi \Vert ^2 +\delta _{\sigma })^{1/2})\le \sigma (\Vert \xi \Vert +\delta _{\sigma }^{{1/2}}),\) for each \(\xi \in \mathbb {R}^n\). Thus, using Remark 3.3 for \(\sigma \) instead of \(\omega \), Leibniz’s rule and the fact that \(\psi \in \mathcal {D}^{\omega }\) we obtain that for each \(\theta >0\) there exist \(\lambda _1, D >0\) such that
for each \(\eta \in \mathbb {N}_0^n.\) Thus, using (2.3), (2.5) and (2.8), there exist \(C_2,\lambda _2>0\) such that
Hence, using (2.3)
Moreover, using Corollary 6.7 there exists \(C>1\) such that,
for \(\epsilon _j=\left[ \frac{\beta _j!}{e^{ [1/(2^j\lambda _2) ]\varphi ^*(2^j\lambda _2 |\beta _j|)}} \right] ^{1/|\beta _j|} \frac{R}{\ell e^{1/(2^j\lambda _2)} C_{\omega }^{1/(2^j\lambda _2|\beta _j|)} },\) where \(j \in \{1, \dots , \ell \}\) and \(C_{\omega }>1\) is such that \(e^{\omega (t)} \le C_{\omega } e^{t}\) (for each \(t\ge 0\)). Observe that using (2.7) it follows
where \(j\in \{1, \dots , \ell \}.\) Thus, \(\frac{R}{\ell e^{1/(2^j\lambda _2)} C_{\omega }^{1/(2^j\lambda _2|\beta _j|)} }<1\) (for \(j\in \{1, \dots , \ell \}\)). Moreover, since \(B(x_0,3R) \subset \Omega \) and \(u\in \mathcal {E}^{\omega }(\Omega ;P)\) there exist \(C_1,\lambda >0\) such that
Hence, putting (6.5), (6.6) and (6.7) together we obtain,
Moreover, using (2.3) and (2.5) we see that denoting \(C_4=C_1C_2\) and \(C_5=\sum _{|\beta |\le m} 1 \) there exist positive constants \(C_6,\lambda _3,\lambda _4\) such that,
Since (6.9) holds true for every \(\ell \in {\mathbb {N}}\) one can take the infimum in \(\ell \) and use [17, Lemma 1.4] to obtain,
Now one can use (\(\gamma \)) to get \(A>0\) such that \(\log \Vert \xi \Vert ^{\rho } \le \frac{\omega (\Vert \xi \Vert ^{\rho })}{2\,m\lambda _3},\) when \(\Vert \xi \Vert >A.\) Thus \(m\log (\Vert \xi \Vert ^{\rho } )-\frac{1}{\lambda _3} \omega (\Vert \xi \Vert ^\rho ) \le -\frac{1}{2\lambda _3} \omega (\Vert \xi \Vert ^\rho ),\) when \(\Vert \xi \Vert >A.\) This, together with the fact that \(0<\rho \le 1\), that \(\omega \) is a increasing and (\(\alpha \))implies that \(\sigma (\Vert \xi \Vert + \delta _{\sigma }^{1/2}) =\omega ( [\Vert \xi \Vert + \delta _{\sigma }^{1/2}]^{\rho })\le \omega ( \Vert \xi \Vert ^{\rho } + \delta _{\sigma }^{\rho /2}) \le \omega (e\Vert \xi \Vert ^\rho ) \le L[\omega (\Vert \xi \Vert ^{\rho })+1],\) when \(\Vert \xi \Vert \ge \frac{\delta _{\sigma }^{1/2}}{(e-1)^{1/\rho }}.\) Hence, choosing \(\theta =\frac{1}{4\lambda _3L}\) it follows that
for \(\Vert \xi \Vert \ge \max \left\{ A;\,\, \frac{\delta }{(e-1)^{1/\rho }} \right\} .\) Therefore, there exists \(C_7>0\) such that
\({(2) \Rightarrow (1):} \) Consider \(0<r<R\), \(\phi \in \mathcal {D}^{\omega }(B(x_0,R))\) (such that \(\phi \equiv 1\) in \(B(x_ 0,r)\)) and \(C,\lambda , c>0\) such that
First, observe that,
from Theorem 1.1 we can conclude that there exists \(0<\delta <r\) such that \(u \in \mathcal {E}^{\sigma }(B(x_0,\delta )).\) In order to prove that \(u \in \mathcal {E}^{\omega }(B(x_0,\delta );P)\) we will consider a compact set \(K\subset B(x_0,\delta )\). Since \(\phi \equiv 1\) in K it follows that \([P(D)]^N(u)(y)= \phi (y) [P(D)]^N(u)(y),\) for each \(y\in K\).
Moreover, given \(v \in {{\mathcal {D}}^{\sigma }}' (\mathbb {R}^n)\) and denoting
it follows from Lemma 5.3 that \(\langle v^{\phi }_j;\psi \rangle \rightarrow \langle v \phi ;\psi \rangle \) (for each \(\psi \in {\mathcal {D}^{\sigma }}(\mathbb {R}^n)\)).
Furthermore for each \(N\in \mathbb {N}_0\) it follows from (6.10) that
Thus,
Hence, using (6.10),
where \(C_1=(2\pi )^{-n} C\int e^{-c \omega (\Vert \xi \Vert ^{\rho })} d\xi \). Therefore,
where \(C_2=C_1|K|\) and \(|K|=\int _{K} 1 dx\) denotes the Lebesgue measure of K, hence concluding that \(\mathcal {E}^{\omega }(B(x_0,{\delta });P).\) \(\square \)
Next, as an immediate consequence of the previous theorem we prove that an iterate of a constant coefficient hypoelliptic operator is a ultradifferentiable function.
Corollary 6.3
Let \(\omega \) be a weight function, \(\Omega \subset \mathbb {R}^n\) be an open set and P(D) be a constant coefficient hypoelliptic linear partial differential operator of order m. Then, denoting \(\sigma (t)=\omega (t^{\rho })\), it follows that \(\mathcal {E}^{\omega }(\Omega ;P)\subset {\mathcal {E}}^{\sigma }(\Omega )\).
Proof
If \(u \in \mathcal {E}^{\omega }(\Omega ;P) \) then, using Theorem 6.2 it follows that u satisfies (6.1). Using Theorem 5.6, for \(N=0\), it follows that \(u \in \mathcal {E}^{\sigma }.\) \(\square \)
The next result is the so called Denjoy-Carleman Kotake-Narasimhan theorem for constant coefficients operator and the proof given here is different from the one in [7].
Corollary 6.4
Let \(\omega \) be a weight function and \(\Omega \subset \mathbb {R}^n\) be an open set. If P(D) is a constant coefficient elliptic linear partial differential operator of order m, then \(\mathcal {E}^{\omega }(\Omega ;P)\subset {\mathcal {E}}^{\omega }(\Omega )\).
Proof
Since P(D) is a constant coefficient elliptic linear partial differential operator it follows that \(\rho =1\) in (1.8). Thus, the result now follows from Corollary 6.3. \(\square \)
Theorem 6.5
Let \(\omega \) be a weight function and \(\Omega \subset \mathbb {R}^n\) be an open set. If P(D) is a constant coefficient linear partial differential operator of order m (non necessarily hypoelliptic), then \(\mathcal E^{\omega }(\Omega )\subset \mathcal {E}^{\omega }(\Omega ;P)\).
Proof
Considering \(u \in \mathcal {E}^{\omega }(\Omega ) \), for each \(K \subset \Omega \) compact there exist \(C,\lambda >0\) such that
for each \(x \in K\) and \(\alpha \in \mathbb {N}_0^n.\) Moreover, denoting \(P(D)= \sum _{|\alpha | \le m} a_{\alpha } D^{\alpha } u\) it follows that
where \(|K|=\int _{K} 1 dx\) denotes the Lebesgue measure of K. Since \(t \mapsto \frac{\varphi ^*(t)}{t} \) is increasing, by denoting \(\displaystyle h=\sum _{|\alpha |\le m} |a_{\alpha }|\) it follows that
Therefore, using (2.5) there exist \(C_1>0\) and \(\lambda _*>0\) such that
Thus we conclude that \(u \in \mathcal {E}^{\omega }(\Omega ;P).\) \(\square \)
Remark 6.6
It is important to observe that using Corollary 6.4 and Theorem 6.5 we can prove that if \(f \in \mathcal {E}^{\omega }(\Omega )\), P(D) is a constant coefficient elliptic linear partial differential operator of order m and \(u \in L^2_{loc}(\Omega )\) is a solution of the equation
then, \(u \in \mathcal {E}^{\omega }(\Omega ).\)
Analogously, we can prove that if P(D) is a non-elliptic operator but it is a constant coefficient hypoelliptic linear partial differential operator of order m satisfying (1.8) and \(u \in L^{2}_{loc}(\Omega )\) satisfies (6.12) then \(u \in {\mathcal {E}}^{\sigma }(\Omega )\) (where \(\sigma (t)=\omega (t^{\rho })\)).
Data Availability
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Notes
A compact set with a finite number of connected components each of which has the property (P) of Whitney, i.e., “there is a number C such that any two points x and y of a connected component L are joined by an arc in L of length less than or equal to \(C|x-y|\)”.
References
Baouendi, M.S., Métivier, G.: Analytic vectors of hypoelliptic operators of principal type. Am. J. Math. 104(2), 287–319 (1982)
Berhanu, S.: A generalization of a microlocal version of Bochner’s theorem Trans. Am. Math. Soc. 374(7), 5269–5285 (2021)
Berhanu, S., Hounie, J.: A class of FBI transforms. Commun. Partial Differ. Equ. 37(1), 38–57 (2012)
Berhanu, S., Cordaro, P., Hounie, J.P.: An Introduction to Involutive Structures. New Mathematical Monographs, 6, Cambridge University Press, Cambridge (2008)
Boiti, C., Jornet, D.: The problem of iterates in some classes of ultradifferentiable functions. Pseudo-differential operators and generalized functions, 21–33, Oper. Theory Adv. Appl., 245, Birkhäuser/Springer, Cham (2015)
Boiti, C., Jornet, D.: A characterization of the wave front set defined by the iterates of an operator with constant coefficients. Rev. R. Acad. Cienc. Exactas Fís. Nat. Ser. A Mat. RACSAM 111(3), 891–919 (2017)
Boiti, C., Jornet, D.: A simple proof of Kotake–Narasimhan theorem in some classes of ultradifferentiable functions. J. Pseudo-Differ. Oper. Appl. 8(2), 297–317 (2017)
Boiti, C., Jornet, D., Juan-Huguet, J.: Wave front sets with respect to the iterates of an operator with constant coefficients. Abstr. Appl. Anal. 17, 438716 (2014)
Bolley, P., Camus, J., Mattera, C.: Analyticité microlocale et itérés d’opérateurs, Séminaire Goulaouic-Schwartz (1978/1979), Exp. No. 13, 9, École Polytech., Palaiseau (1979)
Bolley, P., Camus, J.: Regularité Gevrey et itérés pour une classe d’opérateurs hypoelliptiques. Commun. Partial Differ. Equ. 6(10), 1057–1110 (1981)
Bonet, J., Meise, R., Melikhov, S.N.: A comparison of two different ways to define classes of ultradifferentiable functions. Bull. Belg. Math. Soc. Simon Stevin 14(3), 425–444 (2007)
Braun, R.W., Meise, R., Taylor, B.A.: Ultradifferentiable functions and Fourier analysis. Res. Math. 17, 207–237 (1990)
Christ, M.: Intermediate optimal Gevrey exponents occur. Commun. Partial Differ. Equ. 22(3–4), 359–379 (1997)
Constantine, C.M., Savits, T.H.: A multivariate Faà di Bruno Formula with applications. Trans. Am. Math. Soc. 348(2), 503–520 (1996)
Derridj, M.: On Gevrey vectors of L. Hörmander’s operators. Trans. Am. Math. Soc. 372(6), 3845–3865 (2019)
Derridj, M.: Gevrey regularity of Gevrey vectors of second-order partial differential operators with non-negative characteristic form. Complex Anal. Synerg. 6, 10 (2020)
Fernández, C., Galbis, A., Jornet, D.: Pseudodifferential operators on non-quasianalytic classes of Beurling type. Stud. Math. 167(2), 99–131 (2005)
Hoepfner, G., Medrado, R.: The FBI transforms and their use in microlocal analysis. J. Funct. Anal. 275(5), 1208–1258 (2018)
Hoepfner, G., Rampazo, P.Y.S.: The global Kotake–Narasimhan theorem. Proc. Amer. Math, Soc (2021)
Hörmander, L.: On interior regularity of the solutions of partial differential equations. Commun. Pure Appl. Math. 11, 197–218 (1958)
Hörmander, L.: The analysis of linear partial differential operators II. Springer-Verlag, Berlin-New York (1983)
Hörmander, L.: The Analysis of Linear Partial Differential Operators I. Springer, Berlin (1990)
Juan-Huguet, J.: Iterates and Hypoellipticity of Partial Differential Operators on Non-Quasianalytic Classes. Integr. Equ. Oper. Theory 68, 263–286 (2010)
Juan-Huguet, J.: A Paley–Wiener type theorem for generalized non-quasianalytic classes. Stud. Math. 208(1), 31–46 (2012)
Komatsu, H.: A characterization of Real Analytic Functions Proc. Japan Acad (1960)
Komatsu, H.: Ultradistributions. I. Structure theorems and a characterization. J. Fac. Sci. Univ. Tokyo Sect. IA Math. 20, 25–105 (1973)
Kotake, R., Narasimhan, M.S.: Regularity theorems for fractional powers of a linear elliptic operator. Bull. Soc. Math. France 90, 471 (1962)
Meise, R., Taylor, B.A.: Whitney’s extension theorem for ultradifferentiable functions of Beurling type. Ark. Mat. 26(2), 265–287 (1988)
Nelson, E.: Analytic vectors. Ann. Math. 70, 572–615 (1959)
Newberger, E., Zielezny, Z.: The growth of hypoelliptic polynomials and Gevrey classes. Proc. Am. Math. Soc. 39, 547–552 (1973)
Tartakoff, D.S.: On local Gevrey regularity for Gevrey vectors of subelliptic sums of squares: an elementary proof of a sharp Gevrey Kotake-Narasimhan theorem. Ann. Univ. Ferrara 64, 437–447 (2018)
Treves, F.: Hypo-analytic Structures: Local Theory. Princeton University Press, Princeton (1992)
Funding
G. Hoepfner is supported by FAPESP 2018/14316-3 and CNPq 305955/2021-7. R. Medrado is supported by FAPEAL E:60030.0000002398/2022.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that there is no conflict of interest.
Additional information
Communicated by Dave Walnut.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix: Some Properties for a Constant Coefficient Hypoelliptic Linear Operator
Appendix: Some Properties for a Constant Coefficient Hypoelliptic Linear Operator
Throughout this section we will consider \(x_0 \in \mathbb {R}^n\), \(0<R<1\), \(m>0\) and a constant coefficient hypoelliptic linear operator P(D) of order m, defined in \(B(x_0,R).\) Also, we will denote \(\partial ^{\alpha }P=P^{(\alpha )}\) and \(P^{(\alpha )}(D)f=P^{(\alpha )}f\), for each \(\alpha \in \mathbb {N}_0^n\) and \(f \in C^{\infty }(B(x_0,R)).\) Moreover, in order to simplify the notation we will denote \(\Vert f\Vert _{r}=\Vert f\Vert _{L^2(B(x_0,r))}< \infty ,\) for each \(0<r<R\). In the proof of the next theorem we shall use [20, inequality (4.5)’]. To be more precise, there exist \(C,\gamma >0\) such that,
for each \(\epsilon >0\), \(\phi \in C_c^{m}(B(x_0,R))\) and \(\alpha \in \mathbb {N}_0^n\) (such that \(0<|\alpha |\le m\)). Furthermore, following [20, Lemma 4.1] for each \(\epsilon ,\epsilon _1>0\) such that \(R > \epsilon _1+\epsilon \) there exists \(\phi _{(\epsilon _1,\epsilon )} \in C_c^{\infty } (B(x_0,R-\epsilon ))\) such that \(\phi _{\epsilon _1,\epsilon } \equiv 1\) in \(B(x_0,R-\epsilon _1-\epsilon )\) and there exists \(C>0\) dependent of \(R>0\) but not on \({(\epsilon _1,\epsilon )}\) satisfying,
for \(|\alpha | \le m. \) We will use the following notation,
for \(r\le R,\) \(f \in L^2(B(x_0,r))\) and \(k \in \mathbb {R}.\)
Lemma A.1
Let \(u \in C^{m}(B(x_0,R))\) and \(\gamma >0.\) Then there exists \(D>0\) (dependent of \(\gamma ,P,R\)) such that,
for each \({R>} \delta _1>0\) and \(\alpha \in \mathbb {N}_0^n,\) where \(|\alpha |\le m.\)
Proof
Consider arbitrary \(0<\!\delta \) and \(\delta _1\) such that \(\delta _1+\delta <R.\) By (A.2) there exists \(\phi _{\delta ,{\delta _1}}\!\in C_c^{\infty } \left( B(x_0,R-\delta _1-\tfrac{\delta }{2})\right) ,\) \(\phi _{\delta ,\delta _1} \equiv 1\) in \(B(x_0,R-\delta _1-\delta )\) and
for each \(\alpha \in \mathbb {N}_0^n\) such that \(|\alpha |\le m\). Moreover, for each \(\epsilon >0\) and \(\beta \in \mathbb {N}_0^n\) such that \(|\beta |\le m\) it follows that,
Using (A.1) it follows that,
Next we recall that, \( P(D)(fg)=\sum _{|\alpha |\le m} \frac{1}{\alpha !} P^{(\alpha )}g\,\times \, \partial ^{\alpha } f \) for \(f,g\in C^{\infty }\). Hence,
Thus,
Next, considering an arbitrary \(\chi >0\) and defining \(\epsilon =\frac{\delta }{\chi }\) it follows that,
Since \(0<\delta<R<1\) and \(\gamma >0\) we have \(\delta ^{\gamma }<1\). Considering the supremum in \(\delta >0\) in the above inequality, we have
for each \(\beta \in \mathbb {N}_0^n\) such that \(0<|\beta |\le m\). Summing in \(\beta \) the last inequality we see that, there exists \(C_1>0\) such that
Thus,
Choosing \(\chi >1\), such that \(\chi -C_1 2^{\gamma }>1,\) it follows that,
for each \(\alpha \in \mathbb {N}_0^n\) such that \(0<|\alpha |\le m.\) Therefore, denoting \(C_2=\max \{2^{\gamma };(1+\chi ^{\gamma })\}\) and \(D=C_1 C_2,\) it follows that,
for each \(\alpha \in \mathbb {N}_0^n\) such that \(0<|\alpha |\le m.\) \(\square \)
Corollary 6.7
Let \(u \in C^{m}(B(x_0,R)).\) There exists \(C_P>0\) such that,
for each \(\epsilon ,r>0\) (such that \(r+\epsilon \le R<1\)) and \(\alpha \in \mathbb {N}_0^n\) where \(|\alpha |<m.\) Let us recall the notation \(\Vert u\Vert _{r+\epsilon }=\Vert u\Vert _{L^2(B(x_0,r+\epsilon ))}.\)
Proof
Using Lemma A.1 it follows that,
for each \(\delta ,\delta _1>0\) such that \(\delta +\delta _1<R\) and \(\alpha \in \mathbb {N}_0^n\), such that \(0<|\alpha |\le m.\) Therefore, for each \(0<r<R\) and \(0<\epsilon \le R- r\) it follows that
Hence, using that \(0<R<1\)
where \(C_P=\epsilon ^{\gamma } D\). \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Hoepfner, G., Medrado, R. A New Class of FBI Transforms and Applications. J Fourier Anal Appl 30, 45 (2024). https://doi.org/10.1007/s00041-024-10102-1
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00041-024-10102-1